Monday, September 2, 2013

Catchy Pop Songs?

I've put up a few technical posts lately so time for something more popular. Every now and then I like to look over one of those articles in the mainstream media that is reporting the latest scientific research on music. This is a pretty recent phenomenon as it is only in the last couple of decades that science has paid much attention to music. Some of the research is based on using sophisticated machines to image what is going on in the brain when we listen to music. But other research is more traditional. That seems to be the case with the study that I just ran across today.

Researchers, led by Dr Daniel Mullensiefen of Goldsmiths University, London seem to be compiling their results by "observing thousands of volunteers as they lent their voices to a long list of tunes". It is pretty difficult in these articles in the press to figure out exactly how the results were obtained, but I would speculate that the researchers have no traditional musical training and that the repertoire of songs was extremely arbitrary for some reason. Here is how the Daily Mail describes the research:
Queen's We Are The Champions has been voted the catchiest pop song of all time – by a team of academics.
The scientists observed thousands of volunteers to find out why certain songs inspired unabashed wedding guests and clubbers to belt out their favourites in public.
Singalong hits had four key elements, they concluded: long and detailed musical phrases, multiple pitch changes in a song’s ‘hook’, male vocalists and higher male voices making a noticeable vocal effort.

Top singalongs include Village People’s YMCA, Sum 41’s Fat Lip, Europe’s The Final Countdown and The Automatic’s Monster.
Musical hits rely on ‘maths, science, engineering and technology’, said Dr Daniel Mullensiefen of Goldsmiths University, London.

Researchers solved the karaoke conundrum after observing thousands of volunteers as they lent their voices to a long list of tunes.
There is a somewhat more informative report here. Here is how they went about it:
Singing along to music in a night club is a very cultural activity which at first appears to be quite an erratic behaviour and seems to be mainly governed by coincidental events. One goal of this study that Alisun Pawley from the University of York and I just finished was to find out whether we can find any patterns and regularities in sing-along behaviour using powerful analysis techniques from modern machine learning. And indeed we were able to explain roughly 65% of the variance in the data we collected.
Alisun carried out an under-cover data collection in night clubs across the North of England recoding for each song played the proportion of people singing along to it. She then did a musical analysis of a large subset of songs regarding the vocal performance on the recording as well as the structure of the songs. I then helped with the statistical analysis of the sing-along field data using a small number of contextual variables as well as larger number of musical features.
So what exactly makes a song sing-along-able?
It’s mainly the singer and not the song. If a male singer uses his high chest voice, pronounces the consonants of the lyrics clearly and puts a lot of vocal effort into his performance and if the tune has little vocal embellishment but uses some melodic variation in the hook, then you have a high chance of observing people singing along, especially if it is late at night, if it is a weekend and if the song has been in high up in the UK charts at some point.
To summarize, a researcher visited an undisclosed number of night clubs (five? fifty?) in the north of England for an undisclosed amount of time and collected an undisclosed amount of data about the proportion of people singing along to an undisclosed number of songs. How was this proportion determined? Guesswork? Looks like about half the people here are singing along? Then a "musical analysis" was done of an undisclosed number of songs based on vocal performance and "structure". By the way, as a musician, I have absolutely no idea what could possibly be meant by this description of their "musical analysis". Then a statistical analysis was done (what kind? undisclosed) using a "small number of contextual variables". What were these? Type of alcohol being consumed, clothing worn? Undisclosed. Oh yes and the statistical analysis extended to a "larger" (undisclosed) number of "musical features". What kind of musical features? Undisclosed.

Sounds to me like his assistant Alisun went on a pub run and jotted down some songs and guessed-at proportion of sing-alongers. Then Dr. M├╝llensiefen, with his superior statistical and "maths" skills massaged this into something that vaguely resembles a scientific paper. The results are ... odd. Queen's "We Are the Champions" is a pretty good song and I guess you can sing along with it. The Village People's "YMCA" is even more infectious, but are people still singing along with it? In great numbers? Up there in Yorkshire? Well, ok. But the next ones? "Fat Lip" by Sum 41? Is it even possible to sing along with that? And "The Final Countdown" by Europe? Is anyone even still listening to this gone-to-seed anthem from the 80s? Maybe in Yorkshire...

Here, have a listen and decide for yourself:





My conclusions are that these researchers have no actual musical training (Dr. M├╝llensiefen obtained his doctorate in "the scientific study of music including acoustics, informatics, psychology, neuroscience etc." which in my view means, no, no actual musical training) and that they ended up with some oddly skewed results demonstrating little more than that many people in Yorkshire have appalling musical taste.

What is absolutely typical of this kind of exercise is that it, with no justification whatsoever, generalizes some very limited results to universal truths: "catchiest pop song of all time!" and does so with extremely questionable and mostly undisclosed methods. I guess it sells newspapers.

Now here is a catchy pop song:



2 comments:

Rickard Dahl said...

I agree with you. Doesn't sound much like a proper scientific study considering the lack of data. One important thing in science is that (as far as I know) the experiments should be described so well that someone else can redo the experiment. I study electrical engineering and I doubt a study with this little data would be taken seriously if it would be about an area of physics or engineering. It would be like "we measured this and this but we can't tell you how exactly we did it but here are the results". Besides it's another one of those experiments where scientists with little musical knowledge come to obvious conclusions like "long and detailed musical phrases, multiple pitch changes in a song’s ‘hook’". And what do they mean by multiple pitch changes? To reverse the question why would anyone sing the same note over and over again as a 'hook'?

Bryan Townsend said...

I found so much else to comment on that I never got to those silly musical descriptions. Thanks for the assist! "Long and detailed musical phrases" is one of those expressions that sounds like it might mean something, but of course it means absolutely nothing.

What strange times we live in when you can get a doctorate in "the scientific study of music" and know very little about music.